PROCEDURE FOR PREDICTION OF MULTI-VARIATE TIME SERIES USING BLIND SEPARATION OF INDEPENDENT SOURCES

摘要

The invention relates to a process for prediction of multi-variate time series, with applicability in the fields of engineering, physical sciences, biology, medicine, sociology, hydrology, geophysics, economy, facing measurement and observation data analysis. According to the invention, the process transfers the issue which is the object of analysis from the space of the original data into the space of independent sources, of very much reduced size, where a wide range of methods and techniques are available for modelling and predicting univariate time series, the results of independent source prediction being then transferred back into the original space of the multi-variate time series, using the independent source mixing model obtained from their blind separation.